News 2020

February 2020

Jaime Carbonell Pioneered Use of Computers for Translation

Byron Spice

Jaime Carbonell foresaw a world where people could freely communicate with each other, no matter what language they spoke. He knew that making this dream a reality would require automation, so he spent his career building machines that could understand human language. He knew full well that earlier attempts at machine translation had largely come to naught. Nevertheless, as a young computer science professor at Carnegie Mellon University in 1985, Carbonell persuaded his superiors to let him start a Center for Machine Translation. For the next 35 years, he and his colleagues would create pioneering translation systems and expand the horizons of artificial intelligence. Carbonell, 66, died today following an extended illness. He was the Allen Newell Professor of Computer Science and had earned the distinction of University Professor, the highest academic accolade CMU faculty can attain. Machine translation, which was high-risk research when Carbonell first championed it, is big business today, dominated by tech giants such as Google, Microsoft and Amazon. And the research center he launched would expand in 1996 to become the Language Technologies Institute (LTI), one of seven academic units in CMU's top-ranked School of Computer Science. Under Carbonell's direction, the LTI became the largest and best-known organization of its kind. It has been a leader in areas including natural language processing, question-answering systems, and speech recognition and synthesis, and now boasts five graduate degree programs. "He's kind of the godfather of language technologies," said Raj Reddy, the Moza Bint Nasser University Professor of Computer Science and Robotics and former SCS dean. Carbonell advised more than 60 Ph.D. recipients who continue to spread his influence through academia and industry. But his influence extended well beyond language technologies. Carbonell made many contributions to the field of machine learning, said Tom Mitchell, Founders University Professor of Computer Science. "He played an important and influential role in the early days of machine learning in pulling together researchers working in this area, helping to create an international research community," Mitchell said. His contributions included methods that allowed computers to reason by analogy, developing algorithms that could actively query a teacher to more efficiently learn new concepts, and applying machine learning to diverse real-world applications, ranging from information retrieval and search to protein folding. "Jaime always had astounding levels of energy and creativity," said Robert Frederking, a student of Carbonell's who is now SCS associate dean of doctoral programs. "I have never understood how he could advise maybe a dozen Ph.D. students, run the LTI, personally be the principal investigator on several research projects, teach regularly and travel to DC frequently to work with funding agencies. "And with all that going on," he added, "if you ran a new technical problem by him, he would usually come up with three good suggestions for solution paths to investigate." Carbonell also advised Manuela Veloso, University Professor of Computer Science at CMU, now on leave while she directs AI research at financial services giant J.P. Morgan. She remembers him as an amazing educator and mentor. "With Jaime I learned a lot of AI, but I also learned how to advise," she recalled. "I became a faculty member at Carnegie Mellon and I embraced a lot of what I learned from Jaime. Even now, I still look at Jaime as my advisor, and throughout my career have turned to him for different types of advice. As of now, I have graduated 40 Ph.D. students. I will always thank Jaime for having graduated me." Carbonell grew up in Spanish-speaking Uruguay before his family moved to Boston when he was nine. He earned bachelor's degrees in mathematics and physics at the Massachusetts Institute of Technology and his master's degree and Ph.D. in computer science at Yale University. While studying at MIT, Carbonell worked part-time translating computer manuals into Spanish and developed machine-translation tools to speed up the process. Thus began a fascination with machine translation that would become the focus of his AI research. "We are living in a globalized society, and I think automation has to be the way to overcome language differences," he told the Boston Globe in 1996. Carbonell, who joined CMU in 1979, led teams that developed knowledge-based machine translation of text as well as speech-to-speech translation. He invented a number of well-known algorithms and methods, including maximal marginal relevance (MMR) for summarizing text and a type of machine learning call proactive learning. "Getting the right information to the right people at the right time in the right language in the right medium with the right level of detail" became his mantra. Carbonell played a key role in establishing language technologies as an industry in the Pittsburgh region. "Dr. Carbonell launched, spearheaded, and provided expert advice and support to numerous commercial enterprises, ranging from small startups to Fortune 100 companies," said Alon Lavie, a longtime LTI faculty member who is now vice president of language technologies at Unbabel. Locally, those companies included his own spinoffs, such as Carnegie Speech, Carnegie Group and Wisdom Technologies, and those he advised, such as Vivisimo and Lycos — one of the first successful search engines. Lavie's own spinoff, Safaba Translation Technologies, was acquired by Amazon and became the core of that company's Pittsburgh office. Carbonell created the university's Ph.D. program in language technologies, and is co-creator of the Universal Library and its Million Book Project, which scanned and digitized books and made them available for free online. Carbonell also led a project to apply machine learning techniques to data generated by the aerospace industry and explore how this might improve aircraft maintenance, operating efficiency and reliability. When Carbonell wasn't working, he indulged a lifetime passion for chess. Carbonell is survived by his longtime partner, Yiming Yang; mother, Nelly J. Carbonell, of Concord, Massachusetts; a sister, Ana Maria Carbonell of Berkeley, California; two brothers, Miguel G. Carbonell of Medford, Oregon, and Pablo F. Carbonell of Harvard, Massachusetts; three daughters, Diana Carbonell of Pittsburgh, Isabelle Carbonell of Santa Cruz, California, and Rachel O. Carbonell of Brooklyn, New York; a son, Ruben M. Carbonell, of Southbridge, Massachusetts; and two grandchildren. Funeral arrangements are pending, as are plans for a campus remembrance.

Team Explorer Finishes Second in DARPA Competition

Byron Spice

Team Explorer from Carnegie Mellon University and Oregon State University battled to the end in the DARPA Subterranean Challenge's Urban Circuit competition, ultimately finishing second. Explorer was tied with CoSTAR — a team led by the Jet Propulsion Laboratory that included MIT, CalTech and South Korea's KAIST — going into the last run on the last day of competition on Wednesday. Feb. 26. But CoSTAR located and identified more artifacts during its run, giving it a final score of 16 to Explorer's 11. The Subterranean Challenge, sponsored by the Defense Advanced Research Projects Agency, is a multiyear competition in which autonomous robots map underground environments and look for objects — such as simulated human survivors, backpacks and emergency equipment — much as a first responder might during an emergency response. Explorer deployed various combinations of its three ground robots and four drones during its runs. The CMU Robotics Institute's Sebastian Scherer, who co-leads the team with Matt Travers, a systems scientist, said the performance of the team's drones was a highlight of the competition. "They covered a lot of ground," he said, and worked effectively in air shafts and other tight spaces. Piggybacking and launching the drones from the team's wheeled ground robots also worked well, he added. The team's object-detection methods didn't work so well, said Scherer, an associate teaching professor. One reason was that the unfinished nuclear power plant in which the competition took place is much more cavernous than the vacant hospital where the team practiced before the event. The last run, which pitted the two leading teams head-to-head, was not the only moment of high drama for Explorer. During one of the team's four runs over Urban Circuit courses, one of its wheeled robots flipped completely over while descending stairs, landing on its wheels at the bottom with minimal damage. One of the team's drones, after flying down a shaft, began exploring but flew too low and hit the edge of a door. During Thursday's award ceremony, DARPA even presented the team with a certificate for "Down to the Wire Report," recognizing how it reported an artifact — and earned a point — mere seconds before time expired on its 60-minute run on the course. "You had us on the edge of our seats," said Timothy Chung, DARPA's Subterranean Challenge program manager. In August, Explorer decisively won the challenge's Tunnel Circuit, in which autonomous robots explored mine tunnels. A third event, the Cave Circuit, will take place this August inside natural caves. A final event in August 2021 will include elements of all three subterranean environments. Explorer is one of seven teams that will receive up to $4.5 million from DARPA to develop hardware and software for the competition. It received additional support in the Urban Circuit from Microsoft, Honeywell, Epson, the Richard King Mellon Foundation and CNH Industrial.

Shah Earns NSF CAREER Award

Virginia Alvino Young

Nihar Shah, an assistant professor in the School of Computer Science's Machine Learning and Computer Science Departments, has received a National Science Foundation Faculty Early Career Development (CAREER) Award, the organization's most prestigious award for young faculty members. The five-year, $648,000 award will support his work to improve the fairness of peer-review systems and address issues such as biases and dishonest behavior. Shah's research specialties include machine learning, game theory and crowdsourcing. With the NSF grant, he will design new algorithms for eliciting data from people and processing it in a manner that mitigates biases and unfairness to the greatest possible extent. The research will have a particular focus on peer review of scholarly research, with a wide variety of other applications including hiring, college admissions, recommender systems and crowdsourcing. The project will employ tools from machine learning, statistics, information theory, game theory and social choice. Shah will also release open-source toolkits for practitioners and employ outreach efforts towards creating positive policy change. Shah received his Ph.D. in electrical engineering and computer science from the University of California Berkley in 2017, where his thesis earned the David J. Sakrison Memorial Prize for a "truly outstanding piece of research."

Leading AI Scholars Featured in New Oral Archive

Byron Spice

Artificial intelligence is the creation of human beings, including a number from Carnegie Mellon University. Now, AI is changing humans. It's a subject that Illah Nourbakhsh of the Robotics Institute and Jennifer Keating of the Dietrich College of Humanities and Social Science explore in an interdisciplinary CMU course, "AI and Humanity." Nourbakhsh and Keating have captured the thoughts of some leading AI scholars in a new oral archive that became available online this year. It includes video and transcripts from 22 people, including MIT's Daniela Rus, Harvard University's Barbara Grosz and Microsoft's Eric Horvitz, as well as a number of CMU faculty members such as Martial Hebert, David Danks, Mark Kamlet, Tuomas Sandholm and Jim Herbsleb. "These are the founders of the field," said Nourbakhsh, the K&L Gates Professor of Ethics and Computational Technologies. "It's not a history project; it's about them being critical analysts of their field. They are talking about the future." For instance, Rus, the director of MIT's Computer Science and AI Laboratory, imagines a world where AI and robots take over routine chores. "I like to imagine robots delivering fresh produce at my doorstep on a daily basis," she says in her video. "I like to think about the possibility of garbage cans that take themselves out and automated infrastructure that supports removing them. I like to think about intelligent assistance … that supports us with all sorts of tasks to ensure that we work efficiently, and we live well." Also in the archive, Grosz, a computer scientist at Harvard, suggests people may need to rethink how AI services are designed. Calling a customer service line and talking to a computer first and only to a human if there's a problem makes for a bad experience, she explained. "If they had been designed instead for the person to answer the phone first and for the computer to support them, it would be a better experience all around," she says in her video. Moshe Vardi, a professor of computer science at Rice University, believes people must hold on to those things that make them special, such as human rights. "I don’t think we should give robots human rights," he says in the archive video. "We should not give corporations human rights. We are eroding human rights by giving them away to other things." Nourbakhsh said the plan is to keep adding new voices to the archive, making it an expanding resource for AI students and scholars. The release of the archive coincides with the publication next month by MIT Press of the textbook "AI and Humanity." Nourbakhsh and Keating, assistant dean for educational initiatives in the Dietrich College, wrote it for their course, which is part of Dietrich's Grand Challenge Seminar series for first-year students. The book and the archive will be celebrated on Friday, Feb. 28, in Simmons Auditorium A in the Tepper Building. Vardi, the University Professor and Karen Ostrum George Distinguished Service Professor in Computational Engineering at Rice, will present a keynote lecture, "Technology Is Driving the Future, but Who Is Steering?" at 4 p.m. A panel discussion will follow, featuring Richard Scheines, dean of the Dietrich College; Hebert, dean of the School of Computer Science; Colin MacCabe, professor of English and film at the University of Pittsburgh; and Molly Steenson, K&L Gates Associate Professor of Ethics and Computational Technologies in the College of Fine Arts. The program is open to the public, but an RSVP by Friday, Feb. 21, is requested.

Exploring a Genome's 3D Organization Through a Social Network Lens

Byron Spice

Computational biologists at Carnegie Mellon University have taken an algorithm used to study social networks, such as Facebook communities, and adapted it to identify how DNA and proteins are interconnected into communities within the cell nucleus. Jian Ma, associate professor in CMU's Computational Biology Department, said scientists have come to appreciate that DNA, proteins and other components within the nucleus appear to form structurally and functionally important communities. The behavior of these communities may prove key to understanding basic cellular processes and disease mechanisms, such as aging and cancer development. Figuring out how to identify these communities among the tens of thousands of genes, proteins and other components of the cell is daunting, however. An important factor is proximity — both in terms of genes being controlled by the same regulatory proteins called transcription factors and in terms of spatial arrangement, with the complex folding and packing of DNA putting certain genes close to each other. In many cases, the relationships are similar to many Facebook communities, with some members located near each other, while others who may be far apart are nevertheless drawn together through shared interests. In a paper featured on the cover of the February issue of the journal Genome Research, lead authors Dechao Tian, a post-doctoral researcher, and Ruochi Zhang, a Ph.D. student in computational biology, explain how they developed a new algorithm, MOCHI, to subdivide the interwoven nuclear components into communities. MOCHI was inspired by an algorithm originally developed by the laboratory of computer scientist Jure Leskovec. Beginning as a Ph.D. student at CMU and continuing as a faculty member at Stanford University, Leskovec has specialized in the analysis of large social and information networks. The MOCHI algorithm looks at the spatial arrangement of all the genes and transcription factor proteins in a nucleus based on genome-wide chromosome interactions and global gene regulatory networks. Viewing this information as a 3D graph, the algorithm looks for certain subgraphs or "motifs," within it. A motif might be, say, a triangular shape, as is typical in social network analysis, or a four-node subgraph, which MOCHI uses for analyzing complex networks in the cell nucleus. The algorithm then clusters, or subdivides, the graph in a way that minimizes disruption of these motifs. They tested MOCHI by applying it to five different cell types. Just as the original algorithm has proved adept at identifying communities within a large mass of social network data, MOCHI identified what appear to be hundreds of communities within the nuclei of these cell types. As of yet, the researchers don't know what each community might do, but they say they have reason to believe the subdivisions made by MOCHI are valid. For instance, Ma said that the algorithm identified communities that seem to be common to all of the cell types used in this study. It also identified some communities that appear to be unique to a particular cell type. In addition, Ma said they found "enrichment" of disease related genes within the communities. Much more work will be necessary to identify the function and behavior of each of these communities, Ma said, but the MOCHI algorithm gives researchers a starting point for study. "There's a reason why these communities are formed in the nucleus," he said. "We just don't know the formation mechanisms of these communities yet." Understanding them might help researchers delineate fundamental cellular processes and suggest possible ways to better understand disease development. The researchers also plan to include additional cell nucleus components, such as RNAs and other types of proteins, into their analysis. In addition to Ma, Tian and Zhang, authors of the paper include Yang Zhang and Xiaopeng Zhu, a research associate and a project scientist, respectively, in the Computational Biology Department. The National Institutes of Health, including its 4D Nucleome Program, and the National Science Foundation supported this research.

New Infrastructure Will Enhance Privacy in Today's Internet of Things

Daniel Tkacik

People navigating the digital landscape of today's internet are bombarded with notices about how their data is being collected. But in the physical world — where internet of things (IoT) technologies increasingly track our activities — few, if any, notices are provided. A team of Carnegie Mellon University researchers has created an app and an entire infrastructure to change that. The IoT Privacy Assistant, launched this week, is an app that informs users about what IoT technologies are around them and the data they're collecting. Consider public cameras with facial-recognition and scene-recognition capabilities, Bluetooth beacons tracking your whereabouts at the mall, or your neighbor's smart doorbell. The IoT Privacy Assistant app will let you discover the IoT devices around you and learn about the data they collect. If the device offers choices such as opting in or out of data collection, the app will help you decide. The app is available for both iOS and Android phones. "Because of new laws like the General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA), people need to be informed about what data is collected about them and they need to be given some choices over these processes," said Professor Norman Sadeh, a CyLab faculty member in CMU's Institute for Software Research and the principal investigator on the project. "We have built an infrastructure that enables owners of IoT technologies to comply with these laws, and an app that takes advantage of this infrastructure to empower people to find out about and control data collected by these technologies." Right now, some public spaces under surveillance might have signs that say, "This area is under surveillance," informing people in the vicinity that video of them may be recorded. But Sadeh says this isn't enough. "These signs tell you nothing about what is being done with your footage, how long it's going to be retained, whether or not it uses facial recognition, or with whom this is going to be shared," says Sadeh. "Under regulations like GDPR and CCPA, there are requirements to more explicitly communicate not just the presence of these technologies and what they collect, but also to give people some control over what is being collected and how the data can be used." While end users may access the app to see information about IoT devices around them, owners of IoT devices may use a cloud-based online portal to publish the presence of their IoT devices in registries spanning different areas. Both organizations and individuals can request the creation of registries where they can control the publication of IoT technologies in different areas. The infrastructure is hosted in the cloud and designed to be easy to use. For instance, premade templates for commonly used off-the-shelf IoT devices are available for people to edit and easily publish in registries. "We've done the work for you," Sadeh says. "All you need to do is start adding your IoT resources so you can be in compliance with today's privacy laws." This project has been made possible by a large grant under the Defense Advanced Research Projects Agency's Brandeis privacy research program, as well as through funding from the National Science Foundation's Secure and Trustworthy Cyberspace program. Other members of the team include Yuanyuan Feng, Justin Donnell, Yoshi Torralva, Akshath Jain, Salil Deshpande, and Yaxing Yao.  

Rose and Zimmerman Named AAAS Public Engagement Fellows

Virginia Alvino Young

Two Carnegie Mellon University faculty members have been named Alan I. Leshner Leadership Institute Public Engagement Fellows by the American Association for the Advancement of Science (AAAS). They're part of a cohort of 12 researchers studying artificial intelligence who are being recognized for their demonstrated leadership and research excellence, and their interest in promoting meaningful dialogue between science and society. Carolyn Rose, a professor in the School of Computer Science's Language Technologies Institute and Human-Computer Interaction Institute (HCII), was recognized for a research program that spans AI, education and society. Her work focuses on AI technologies including machine learning, text mining, conversational agents and robotics. John Zimmerman is the Tang Family Professor of Artificial Intelligence and Human-Computer Interaction in the HCII. He was recognized for his research on interaction with intelligent systems, service innovation via social computing, human-robot interaction and how technical systems can help people become the person they desire to be. The Leshner Leadership Institute is an initiative of the AAAS Center for Public Engagement With Science and Technology. This year's cohort of Public Engagement Fellows focusing on AI will convene in June for a week of intensive public engagement and science communication training and public engagement plan development.

Carnegie Mellon Will Compete in Abu Dhabi Robotics Challenge

Byron Spice

A group of Carnegie Mellon University students and faculty is among 30 teams from around the world that will compete at the biennial Mohamed Bin Zayed International Robotics Challenge (MBZIRC), one of the world’s largest and most prestigious robotics competitions. The event will take place Feb. 23–25. The competition in the United Arab Emirates will literally be a trial by fire, said Oliver Kroemer, co-leader of CMU’s Tartan team, as the organizers have advised that one segment of the competition — a firefighting challenge — will include real flames. "The competition includes a nice suite of forward-looking applications," said Kroemer, an assistant professor in CMU’s Robotics Institute. The firefighting challenge, for instance, will use drones and ground-based robots to detect and extinguish high-rise fires. Other challenges include a construction scenario in which robot teams place large blocks in a row. In another, addressing the threat of drones flying too close to airports and military installations, teams will use their drones to track and intercept aerial targets. The Tartan team includes a number of students enrolled in the Robotics Institute’s master’s program in systems development, as well as some undergraduates from the University of Pittsburgh. Kroemer and eight of the students will make the trip to the competition, which is organized and hosted by Khalifa University of Science and Technology in Abu Dhabi. "Building on the success of the inaugural edition of MBZIRC 2017, the second edition will bring a total of 30 teams to the UAE to display their advanced innovations in some key robotics technologies," said Arif Sultan Al Hammadi, executive vice president of Khalifa University. "These finalists represent top internationally renowned academic and research institutions with well-established robotics labs, and we believe MBZIRC 2020 will showcase the creative best in robotics and stand testimony to its reputation as a leading global competition." One of the attractions of the competition is that tackling the tasks in each challenge requires that teams build systems, a CMU strongpoint, Kroemer said. "Plus, there’s just the fun factor," he added. "Students enjoy the competition." Though the Tartans will field one ground robot, the competition relies heavily on drones. The robots will operate autonomously during the competition, with teams charged demerits if people need to intervene. Individual competitions will occur for each of the three challenges — drone safety, construction and firefighting — as well as a grand challenge that is a triathlon encompassing all of the three individual challenges. Winners of each individual challenge will receive a prize of one million UAE dirhams, equivalent to about $270,000. The grand challenge winner receives two million dirhams. Some of the world’s top academic robotics programs will be represented at the three-day competition, among them Georgia Tech, Virginia Tech, Korea Advanced Institute of Science and Technology, China’s Tsinghua University, and the University of Tokyo. Teams that were champions at MBZIRC 2017, including the University of Bonn, University of Pennsylvania and Beijing University of Technology, will compete once again. The Tartans were part of the inaugural contest, but competed in only one of the challenges. The competition coincides with another high-profile robotics contest, the Defense Advanced Research Project Agency (DARPA) Subterranean Challenge, in which CMU also fields a team, Explorer. Sebastian Scherer, an associate research professor in the Robotics Institute, is co-leader of both CMU’s MBZIRC entry and the DARPA entry. The DARPA competition, which requires teams of robots to perform search-and-rescue scenarios underground, will take place Feb. 18–27 in Elma, Washington.  

Team Explorer Adds Capabilities for Latest DARPA Robotics Contest

Byron Spice

Maneuvering wheeled robots up and down stairwells and flying drones slim enough to slip through narrow doorways and tough enough to survive collisions are among the new capabilities Team Explorer has added for the latest competition in the DARPA Subterranean Challenge.Explorer, which includes researchers from Carnegie Mellon University and Oregon State University, is one of 11 teams that will send robots into the depths of an incomplete nuclear power plant in a search-and-rescue scenario. Teams have been told to be ready to search multiple levels and open spaces within the plant, looking for artifacts such as a simulated human survivor, a backpack, a cellphone, a vent and a mock gas leak.The Urban Circuit is the second in a series of scored contests in the Defense Advanced Research Projects Agency's (DARPA) Subterranean Challenge, in which autonomous robots map underground environments and look for objects, much as a first responder might in an emergency. Teams are scored based on the number of artifacts that the robots accurately identify and map.The Urban Circuit takes place Feb. 18–27 in Elma, Washington.Explorer outscored the other competitors during the first contest, the Tunnel Circuit, this past August. But Matt Travers, who co-leads the team with Sebastian Scherer, said performance in coal mine tunnels doesn't necessarily predict success in the very different environment of a power plant."Searching across several levels and in a variety of spaces will require new equipment and even some new search tactics," he said. "In a tunnel, the artifacts are all fairly close to where our robots are driving anyway. In a plant, we're less constrained, so finding things may be more difficult."Stairways are also a big challenge."Stairs are right at the upper limit of what's possible with a wheeled vehicle," said Travers, a systems scientist in CMU's Robotics Institute. Two of the team's wheeled robots each weigh close to 500 pounds, so getting them down a stairwell without flipping over is a feat. A third vehicle, weighing half as much but with the same power, has been added to the team's fleet because it can climb stairs.But navigating stairs autonomously is not yet possible, said Steve Willits, the team's lead test engineer. The robots will be teleoperated on stairs because it is important to get and keep the robots properly aligned on the stairs.In the previous contest, the robots had cameras that could transmit snapshots of artifacts to help the human operator, who was located on the surface, identify them. But teleoperation required that the team beef up the communications system to enable streaming video, Willits said. The streaming video also will provide a secondary means of verifying artifacts.Changes also have been made to the team's drones."From the ground up, they've been designed to fly indoors," said Scherer, an associate research professor in the Robotics Institute. Each craft's four rotors, usually on the top corners of a drone, have been moved underneath and pushed closer together, so the drones are just 26 inches wide.The drones also have been equipped with bumpers, anticipating that they will need to survive bumps and other collisions with walls, doorways, posts and equipment. Because they run out of power faster than ground vehicles, the drones will be carried on the backs of the wheeled robots until they need to be deployed.Willits said the ground robots will make 360-degree turns before launching the drones, using their cameras to ensure that they're not near a pole or piece of equipment that a drone might strike when it becomes airborne.In addition to Scherer and Travers, Geoff Hollinger, an associate professor of mechanical engineering at Oregon State, leads a team of researchers who specialize in coordinating multiple-robot teams. Explorer now has a fleet of seven robots, four or five of which will be deployed at any given time, said Chao Cao, a systems engineer in the CMU Robotics Institute who will be the team's lone operator during the competition.Team Explorer is sponsored in the Urban Circuit by Microsoft, Honeywell, Epson, the Richard King Mellon Foundation and CNH Industrial.

Chouldechova and Zhu Awarded Research Grant on Fairness in Artificial Intelligence

Virginia Alvino Young (SCS), Caitlin Kizielewicz (Heinz)

The National Science Foundation (NSF) and Amazon have announced that two Carnegie Mellon University researchers have been awarded a research grant as part of the NSF's Program on Fairness in Artificial Intelligence (AI), in collaboration with Amazon. Alexandra Chouldechova is an assistant professor of statistics and public policy at the Heinz College of Information Systems and Public Policy. Haiyi Zhu is an assistant professor in the School of Computer Science's Human Computer Interaction Institute (HCII). As recipients of the grant, Chouldechova and Zhu will serve as co-principal investigators of a project that will focus on advancing AI with human-algorithm collaborations. In collaboration with Steven Wu, assistant professor of computer science and engineering at the University of Minnesota, and Min Kyung Lee, assistant professor at the University of Texas at Austin School of Information, the research will aim to close the gap between algorithmic objectives and the complications of real-world decision-making contexts. "It is an honor to receive this award from the National Science Foundation and Amazon," Chouldechova said. "AI and machine learning are increasingly becoming parts of everyday life, and I look forward to contributing to the research that will influence how AI is used for years to come." In 2017, Amazon approached the NSF about creating a partnership to support research in AI and fairness. As a result, the Program on Fairness was born. Two years later, the NSF released a joint solicitation for research proposals. The award granted to Chouldechova is among those issued in the first-ever cohort and is anticipated to run for two additional years. Projects funded under this joint program are expected to result in open-source tools, publicly available datasets and widely accessible publications. These deliverables stand to benefit academia, government agencies and private companies. "We work closely with stakeholders to define fairness, navigate fairness-utility trade-offs and understand the fairness of AI in the real world," Zhu said. The project will also account for the cognitive strengths and limitations of human decision-makers throughout the development and deployment of the algorithmic system. As Estella Loomis McCandless Assistant Professor of Statistics and Public Policy at Heinz College, Chouldechova's teaching and research focus on problems related to fairness in predictive modeling. Chouldechova is a graduate of the University of Toronto and received her Ph.D. in statistics from Stanford University. Haiyi Zhu is the Daniel P. Siewiorek Assistant Professor of Human-Computer Interaction in CMU's HCII. She received a B.S in computer science from Tsinghua University and an M.S. and Ph.D. in human-computer interaction from CMU. Her research interests lie at the intersection of human-computer interaction, machine learning and organization science.

Kolter, Gkioulekas Named Sloan Research Fellows

$story.metadata.author

Zico Kolter, an associate professor in the Computer Science Department (CSD), and Ioannis Gkioulekas, an assistant professor in the Robotics Institute, are among 126 early career researchers to receive 2020 Sloan Research Fellowships. The prestigious fellowships honor outstanding scholars in the U.S. and Canada in eight scientific and technical fields: chemistry, computer science, economics, mathematics, computational and evolutionary molecular biology, neuroscience, ocean sciences, and physics. "To receive a Sloan Research Fellowship is to be told by your fellow scientists that you stand out among your peers," said Adam F. Falk, president of the Alfred P. Sloan Foundation. "A Sloan Research Fellow is someone whose drive, creativity and insight make them a researcher to watch." Kolter, who joined Carnegie Mellon University in 2012, is a leading expert in artificial intelligence research. He develops methods that make machine learning more robust, interpretable and modular. He also has worked on applications for smart energy and sustainability solutions. In addition to his full-time post in CSD, Kolter is chief scientist of AI research for the Bosch Center for AI, working out of its Pittsburgh office. "Zico is one of the top machine learning researchers of his generation," said Ariel Procaccia, the Gordon McKay Professor of Computer Science at Harvard University, who nominated Kolter for the fellowship. "He's made fundamental contributions to deep learning, computational sustainability and robotics. He has an amazing ability to put theoretical insights into practice, often building on his absolute mastery of optimization." Procaccia received a Sloan fellowship while a CSD faculty member in 2015. Gkioulekas, who joined CMU in 2017, is a leading researcher in computational imaging, computer vision and computer graphics. Most recently, he has focused on non-line-of-sight imaging. In this emerging field, researchers reconstruct images from scattered light, enabling them to see objects not normally visible because they are located around corners or behind gauzy filters. He also is part of a National Science Foundation Expeditions in Computing program to develop special cameras that can see through skin to diagnose and monitor health conditions at a cellular scale. "Ioannis is redefining the concept of a camera by creating breakthroughs in inverse algorithms applied to non-line-of-sight imaging and imaging below the skin," said Srinivasa Narasimhan, interim director of the Robotics Institute, who nominated Gkioulekas for the fellowship. Nearly a thousand researchers are nominated each year for 126 Sloan fellowship slots. Winners receive a two-year, $75,000 fellowship that can be spent to advance their research. To date, 60 CMU faculty members have received Sloan fellowships since they were first awarded in 1955.  

Ranysha Ware Awarded Prize for Work on Internet Fairness

Virginia Alvino Young

Ranysha Ware, a Ph.D. student in Carnegie Mellon University’s Computer Science Department, has received a 2020 Applied Networking Research Prize (ANRP) from the Internet Engineering Task Force. She is being recognized for her work on congestion control fairness. Ware leads a research project on internet fairness that recently demonstrated how Google’s new congestion control algorithm (CCA) gives an unfair advantage to its own traffic, and proposed new guidelines for developing future algorithms. She is one of six recipients of the ANRP this year, and one of two who will present their work March 21–27 at the IETF 107 in Vancouver, British Columbia. Ware earned her M.S. in computer science from the University of Massachusetts Amherst. She is a Facebook Emerging Scholar and two-time recipient of the National GEM Consortium Fellowship.

New Software Agents Will Infer What Users Are Thinking

Byron Spice

Personal assistants today can figure out what you are saying, but what if they could infer what you were thinking based on your actions? A team of academic and industrial researchers led by Carnegie Mellon University is working to build artificially intelligent agents with this social skill. The goal of the four-year, $6.6 million project, led by Katia Sycara of CMU's Robotics Institute and sponsored by the Defense Advanced Research Projects Agency, is to use machine social intelligence to help human-and-machine teams work together safely, efficiently and productively. The project includes human factors experts and neuroscientists at the University of Pittsburgh and Northrop Grumman. "The idea is for the machine to try to infer what people are thinking based on their behavior," said Sycara, a research professor who has spent decades developing software agents. "Is the person confused? Are they paying attention to what is needed? Are they overloaded?" In some cases, the software agent might even be able to determine that a teammate is making mistakes because of misinformation or lack of training, she added. Humans have the ability to infer the mental states of others, called theory of mind — something people do as part of situational awareness while evaluating their environment and considering possible actions. AI systems aren't yet capable of this skill, but Sycara and her colleagues expect to achieve this through meta-learning, a branch of machine learning in which the software agent essentially learns how to learn. In addition to Sycara, the team includes three co-principal investigators: Changliu Liu, an assistant professor in the Robotics Institute; Michael Lewis, a professor at Pitt's School of Computing and Information; and Ryan McKendrick, a cognitive scientist at Northrop Grumman. The research team will test their socially intelligent agents in a search-and-rescue scenario within the virtual world of the Minecraft video game, in a testbed developed with researchers at Arizona State University. In the first year, the researchers will focus on training their software agent to infer the state of mind of an individual team member. In subsequent years, the agent will interact with multiple human players and attempt to understand what each of them is thinking, even as their virtual environment changes. Software agents are autonomous programs that can perceive their environment and make decisions. In addition to digital personal assistants, examples range from the programs that operate self-driving cars to those that cause advertisements to pop up in emails for products in which the user has expressed interest. Agents also can be used to help people with complex tasks, such as scheduling and logistics. DARPA is sponsoring the project through its Artificial Social Intelligence for Successful Teams (ASIST) program.

Asakawa Wins Helen Keller Achievement Award

Byron Spice

The American Foundation for the Blind (AFB) today announced that Chieko Asakawa, an IBM Fellow and the IBM Distinguished Service Professor in Carnegie Mellon University's Robotics Institute, is one of three recipients of its prestigious Helen Keller Achievement Awards this year. The AFB award recognizes Asakawa for her life's work in furthering accessibility research and development, from her digital braille work in the 1980s to the screen-reading software of the IBM Home Page Reader in the 1990s. Most recently, she has collaborated with CMU researchers to develop artificial intelligence and robotic technologies that make urban environments — such as shopping malls, airports and university campuses — accessible to people who are blind. Other award recipients include the American Printing House for the Blind, for its ongoing commitment to accessibility and a future that belongs to everyone; and Procter & Gamble, for prioritizing inclusive design and an inclusive and diverse work environment. The honorees will receive their awards at a ceremony on Wednesday, March 25, in Arlington, Virginia. "All of our honorees exemplify the spirit of Helen Keller in their efforts to create a more inclusive and rewarding quality of life for people who are blind or visually impaired," said AFB President and CEO Kirk Adams. "The examples set by our longtime friends and partners at the American Printing House for the Blind, Dr. Asakawa, and Procter & Gamble illustrate that the responsibility of building a future of no limits for people with disabilities is everyone's, and we are proud to recognize their outstanding work." Asakawa, who has been blind since age 14, came to CMU in 2014 and has led the development of NavCog, a smartphone app that analyzes signals from Bluetooth beacons to help people with visual disabilities navigate their surroundings. The system has been deployed on the CMU campus, in a Japanese shopping mall and at Pittsburgh International Airport. She also is working on "smart suitcase" technology, including BBeep, a rolling suitcase with cameras that helps blind people clear a path in crowds and avoid collisions. Asakawa has received numerous honors for her accessibility research. Last year, she was inducted into the National Inventors Hall of Fame. In 2013, she received the Medal of Honor with Purple Ribbon from the government of Japan.

Filter News

News by Year